Shreyash Arya

I'm an AI Research Engineer at Detesia GmbH and a Research Collaborator at the Max Planck Institute for Informatics (MPI-INF), working in the Computer Vision and Machine Learning group (D2) with Prof. Dr. Bernt Schiele. I completed my Master's in Computer Science at Saarland University on the Saarland Informatics Campus, where my thesis focused on increasing interpretability of deep neural networks via B-cosification.

Previously, I was a Research Assistant at MPI-INF and the German Research Center for Artificial Intelligence (DFKI) in Saarbrücken, and a Researcher at TCS Research and Innovation Labs in New Delhi. My research focuses on Explainable AI (xAI), Interpretability in Computer Vision, and Deepfake Detection.

Email  /  Google Scholar  /  Github  /  LinkedIn  /  X  /  Bluesky  /  ResearchGate

Paper | Code | Talk @ Cohere for AI | neptune.ai Interview

Paper

Paper

Paper

Paper

• 2025: Presenting "B-cosification" at DAGM GCPR 2025, Freiburg, Germany
• 2025: Presenting "B-cosification" at ICVSS 2025, Sicily, Italy
• 2025: Presenting "B-cosification" at CVPR@Paris 2025
• 2024: Invited Talk at Cohere for AI on "B-cosification: Transforming Deep Neural Networks to be Inherently Interpretable"
• 2024: Video Interview with neptune.ai on B-cosification

• M.Sc. Computer Science, Saarland University, Saarland Informatics Campus, Saarbrücken, Germany
  Thesis: "Increasing Interpretability of Deep Neural Networks via B-cosification"
  Supervised by Prof. Dr. Bernt Schiele, Max Planck Institute for Informatics
• B.Tech. Computer Science and Engineering, IIIT-Delhi, New Delhi, India

• AI Research Engineer, Detesia GmbH, Saarbrücken — ML Explainability for Deepfake Detection
• Research Collaborator, Max Planck Institute for Informatics — Computer Vision and Machine Learning (D2) with Prof. Dr. Bernt Schiele
• Research Assistant, German Research Center for Artificial Intelligence (DFKI) — Cognitive Assistants, MePheSTO project with Inria, France
• Researcher, TCS Research and Innovation Labs, New Delhi

• Privacy Attacks in Machine Learning — Membership Inference, Attribute Inference and Model Inversion attacks implemented using PyTorch
• Pokemon Diffusion — Generating new Pokemons using Deep Generative Diffusion Models (DDPM)
• Conv-AcT — Human Activity Recognition (HAR) with Vision Transformer (ViT) based on Convolutional Features
• Adversarial Defense — Proactive and reactive measures using PCL loss and image transformations with adversarial example transferability
• Personified Chatbot — A chatbot responding based on the answering pattern of Dr. APJ Abdul Kalam using IR, NLP, and Deep Learning

Template from Jon Barron.